Effect of Confidence Indicators on Trust in AI-Generated Profiles

Bruzzese, Tommy; Gao, Irena; Dietz, Griffin; Ding, Christina; Romanos, Alyssa B.

doi:10.1145/3334480.3382842

Cited by 5 publications

(2 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These responses suggest that efforts to flag bot content may mitigate their ability to spread misinformation on Twitter. These results align with previous studies finding that people consider identified or suspected bots as less trustworthy than human-generated content (Waddell 2018 ; Bruzzese et al 2020 ; Graefe and Bohlken 2020 ; Jakesch et al 2019 ). Additionally, because “the author is the feature that led the users to the most accurate perceptions” about the veracity of information (Zubiaga and Ji 2014 ), it is not surprising that finding out that Twitter posters are bots reduces user attitudes about the tweet.…”

Section: Discussionsupporting

confidence: 91%

“…Readers did not assign higher credibility scores to human-written vs. bot-written news articles when they did not know who wrote the story, but they considered stories labeled as written by humans more credible and readable (Graefe and Bohlken 2020 ). Adding low-confidence indicators to AI-generated content decreases participant trust, but high-confidence indicators do not increase trust (Bruzzese et al 2020 ). Research of participants who viewed tweets labeled as coming from either a CDC Twitterbot or a human working at the CDC found that “a Twitterbot is perceived as a credible source of information” (Edwards et al 2014 , p. 374).…”

Section: Literature Reviewmentioning

confidence: 99%

See 1 more Smart Citation

Use of bot and content flags to limit the spread of misinformation among social networks: a behavior and attitude survey

Lanius

Weber

MacKenzie

2021

Soc. Netw. Anal. Min.

View full text Add to dashboard Cite

The COVID-19 infodemic is driven partially by Twitter bots. Flagging bot accounts and the misinformation they share could provide one strategy for preventing the spread of false information online. This article reports on an experiment ( N = 299) conducted with participants in the USA to see whether flagging tweets as coming from bot accounts and as containing misinformation can lower participants’ self-reported engagement and attitudes about the tweets. This experiment also showed participants tweets that aligned with their previously held beliefs to determine how flags affect their overall opinions. Results showed that flagging tweets lowered participants’ attitudes about them, though this effect was less pronounced in participants who frequently used social media or consumed more news, especially from Facebook or Fox News. Some participants also changed their opinions after seeing the flagged tweets. The results suggest that social media companies can flag suspicious or inaccurate content as a way to fight misinformation. Flagging could be built into future automated fact-checking systems and other misinformation abatement strategies of the social network analysis and mining community.

show abstract

Section: Discussionsupporting

confidence: 91%

Section: Literature Reviewmentioning

confidence: 99%